Merge remote-tracking branch 'upstream/main' into aa/getline_precond

Author: alejandro-alvarez-sonarsource

.github/workflows/issue-release-workflow.yml

@@ -65,5 +65,5 @@ jobs:
           release-workflow \
           --branch-repo-token ${{ secrets.RELEASE_WORKFLOW_PUSH_SECRET }} \
           --issue-number ${{ github.event.issue.number }} \
-          --requested-by ${{ github.event.issue.user.login }} \
+          --requested-by ${{ (github.event.action == 'opened' && github.event.issue.user.login) || github.event.comment.user.login }} \
           auto

.github/workflows/release-binaries.yml

@@ -35,6 +35,7 @@ jobs:
   prepare:
     name: Prepare to build binaries
     runs-on: ubuntu-22.04
+    if: github.repository == 'llvm/llvm-project'
     outputs:
       release-version: ${{ steps.vars.outputs.release-version }}
       flags: ${{ steps.vars.outputs.flags }}
@@ -47,11 +48,16 @@ jobs:
     - name: Checkout LLVM
       uses: actions/checkout@b4ffde65f46336ab88eb53be808477a3936bae11 # v4.1.1
 
+    - name: Install Dependencies
+      run: |
+        pip install -r ./llvm/utils/git/requirements.txt
+
     - name: Check Permissions
       env:
         GITHUB_TOKEN: ${{ github.token }}
+        USER_TOKEN: ${{ secrets.RELEASE_TASKS_USER_TOKEN }}
       run: |
-        ./llvm/utils/release/./github-upload-release.py --token "$GITHUB_TOKEN" --user ${{ github.actor }} check-permissions
+        ./llvm/utils/release/./github-upload-release.py --token "$GITHUB_TOKEN" --user ${{ github.actor }} --user-token "$USER_TOKEN" check-permissions
 
     - name: Collect Variables
       id: vars
@@ -85,6 +91,7 @@ jobs:
     name: "Fill Cache ${{ matrix.os }}"
     needs: prepare
     runs-on: ${{ matrix.os }}
+    if: github.repository == 'llvm/llvm-project'
     strategy:
       matrix:
         os:
@@ -119,6 +126,7 @@ jobs:
       - prepare
       - fill-cache
     runs-on: ${{ matrix.target.runs-on }}
+    if: github.repository == 'llvm/llvm-project'
     strategy:
       fail-fast: false
       matrix:

bolt/lib/Rewrite/LinuxKernelRewriter.cpp

@@ -27,6 +27,21 @@ using namespace bolt;
 
 namespace opts {
 
+static cl::opt<bool>
+    AltInstHasPadLen("alt-inst-has-padlen",
+                     cl::desc("specify that .altinstructions has padlen field"),
+                     cl::init(false), cl::Hidden, cl::cat(BoltCategory));
+
+static cl::opt<uint32_t>
+    AltInstFeatureSize("alt-inst-feature-size",
+                       cl::desc("size of feature field in .altinstructions"),
+                       cl::init(2), cl::Hidden, cl::cat(BoltCategory));
+
+static cl::opt<bool>
+    DumpAltInstructions("dump-alt-instructions",
+                        cl::desc("dump Linux alternative instructions info"),
+                        cl::init(false), cl::Hidden, cl::cat(BoltCategory));
+
 static cl::opt<bool>
     DumpExceptions("dump-linux-exceptions",
                    cl::desc("dump Linux kernel exception table"),
@@ -157,6 +172,9 @@ class LinuxKernelRewriter final : public MetadataRewriter {
   /// Alignment of paravirtual patch structures.
   static constexpr size_t PARA_PATCH_ALIGN = 8;
 
+  /// .altinstructions section.
+  ErrorOr<BinarySection &> AltInstrSection = std::errc::bad_address;
+
   /// Section containing Linux bug table.
   ErrorOr<BinarySection &> BugTableSection = std::errc::bad_address;
 
@@ -205,6 +223,9 @@ class LinuxKernelRewriter final : public MetadataRewriter {
 
   Error readBugTable();
 
+  /// Read alternative instruction info from .altinstructions.
+  Error readAltInstructions();
+
   /// Mark instructions referenced by kernel metadata.
   Error markInstructions();
 
@@ -232,6 +253,9 @@ class LinuxKernelRewriter final : public MetadataRewriter {
     if (Error E = readBugTable())
       return E;
 
+    if (Error E = readAltInstructions())
+      return E;
+
     return Error::success();
   }
 
@@ -476,7 +500,8 @@ Error LinuxKernelRewriter::readORCTables() {
     // Consume the status of the cursor.
     if (!IPCursor)
       return createStringError(errc::executable_format_error,
-                               "out of bounds while reading ORC IP table");
+                               "out of bounds while reading ORC IP table: %s",
+                               toString(IPCursor.takeError()).c_str());
 
     if (IP < PrevIP && opts::Verbosity)
       BC.errs() << "BOLT-WARNING: out of order IP 0x" << Twine::utohexstr(IP)
@@ -498,7 +523,8 @@ Error LinuxKernelRewriter::readORCTables() {
     // Consume the status of the cursor.
     if (!ORCCursor)
       return createStringError(errc::executable_format_error,
-                               "out of bounds while reading ORC");
+                               "out of bounds while reading ORC: %s",
+                               toString(ORCCursor.takeError()).c_str());
 
     if (Entry.ORC == NullORC)
       continue;
@@ -819,7 +845,8 @@ Error LinuxKernelRewriter::readStaticCalls() {
     // Consume the status of the cursor.
     if (!Cursor)
       return createStringError(errc::executable_format_error,
-                               "out of bounds while reading static calls");
+                               "out of bounds while reading static calls: %s",
+                               toString(Cursor.takeError()).c_str());
 
     ++EntryID;
 
@@ -930,8 +957,10 @@ Error LinuxKernelRewriter::readExceptionTable() {
 
     // Consume the status of the cursor.
     if (!Cursor)
-      return createStringError(errc::executable_format_error,
-                               "out of bounds while reading exception table");
+      return createStringError(
+          errc::executable_format_error,
+          "out of bounds while reading exception table: %s",
+          toString(Cursor.takeError()).c_str());
 
     ++EntryID;
 
@@ -1037,8 +1066,10 @@ Error LinuxKernelRewriter::readParaInstructions() {
     const uint8_t Len = DE.getU8(Cursor);
 
     if (!Cursor)
-      return createStringError(errc::executable_format_error,
-                               "out of bounds while reading .parainstructions");
+      return createStringError(
+          errc::executable_format_error,
+          "out of bounds while reading .parainstructions: %s",
+          toString(Cursor.takeError()).c_str());
 
     ++EntryID;
 
@@ -1105,7 +1136,8 @@ Error LinuxKernelRewriter::readBugTable() {
 
     if (!Cursor)
       return createStringError(errc::executable_format_error,
-                               "out of bounds while reading __bug_table");
+                               "out of bounds while reading __bug_table: %s",
+                               toString(Cursor.takeError()).c_str());
 
     ++EntryID;
 
@@ -1132,6 +1164,125 @@ Error LinuxKernelRewriter::readBugTable() {
   return Error::success();
 }
 
+/// The kernel can replace certain instruction sequences depending on hardware
+/// it is running on and features specified during boot time. The information
+/// about alternative instruction sequences is stored in .altinstructions
+/// section. The format of entries in this section is defined in
+/// arch/x86/include/asm/alternative.h:
+///
+///   struct alt_instr {
+///     s32 instr_offset;
+///     s32 repl_offset;
+///     uXX feature;
+///     u8  instrlen;
+///     u8  replacementlen;
+///	    u8  padlen;         // present in older kernels
+///   } __packed;
+///
+/// Note the structures is packed.
+Error LinuxKernelRewriter::readAltInstructions() {
+  AltInstrSection = BC.getUniqueSectionByName(".altinstructions");
+  if (!AltInstrSection)
+    return Error::success();
+
+  const uint64_t Address = AltInstrSection->getAddress();
+  DataExtractor DE = DataExtractor(AltInstrSection->getContents(),
+                                   BC.AsmInfo->isLittleEndian(),
+                                   BC.AsmInfo->getCodePointerSize());
+  uint64_t EntryID = 0;
+  DataExtractor::Cursor Cursor(0);
+  while (Cursor && !DE.eof(Cursor)) {
+    const uint64_t OrgInstAddress =
+        Address + Cursor.tell() + (int32_t)DE.getU32(Cursor);
+    const uint64_t AltInstAddress =
+        Address + Cursor.tell() + (int32_t)DE.getU32(Cursor);
+    const uint64_t Feature = DE.getUnsigned(Cursor, opts::AltInstFeatureSize);
+    const uint8_t OrgSize = DE.getU8(Cursor);
+    const uint8_t AltSize = DE.getU8(Cursor);
+
+    // Older kernels may have the padlen field.
+    const uint8_t PadLen = opts::AltInstHasPadLen ? DE.getU8(Cursor) : 0;
+
+    if (!Cursor)
+      return createStringError(
+          errc::executable_format_error,
+          "out of bounds while reading .altinstructions: %s",
+          toString(Cursor.takeError()).c_str());
+
+    ++EntryID;
+
+    if (opts::DumpAltInstructions) {
+      BC.outs() << "Alternative instruction entry: " << EntryID
+                << "\n\tOrg:     0x" << Twine::utohexstr(OrgInstAddress)
+                << "\n\tAlt:     0x" << Twine::utohexstr(AltInstAddress)
+                << "\n\tFeature: 0x" << Twine::utohexstr(Feature)
+                << "\n\tOrgSize: " << (int)OrgSize
+                << "\n\tAltSize: " << (int)AltSize << '\n';
+      if (opts::AltInstHasPadLen)
+        BC.outs() << "\tPadLen:  " << (int)PadLen << '\n';
+    }
+
+    if (AltSize > OrgSize)
+      return createStringError(errc::executable_format_error,
+                               "error reading .altinstructions");
+
+    BinaryFunction *BF = BC.getBinaryFunctionContainingAddress(OrgInstAddress);
+    if (!BF && opts::Verbosity) {
+      BC.outs() << "BOLT-INFO: no function matches address 0x"
+                << Twine::utohexstr(OrgInstAddress)
+                << " of instruction from .altinstructions\n";
+    }
+
+    BinaryFunction *AltBF =
+        BC.getBinaryFunctionContainingAddress(AltInstAddress);
+    if (AltBF && BC.shouldEmit(*AltBF)) {
+      BC.errs()
+          << "BOLT-WARNING: alternative instruction sequence found in function "
+          << *AltBF << '\n';
+      AltBF->setIgnored();
+    }
+
+    if (!BF || !BC.shouldEmit(*BF))
+      continue;
+
+    if (OrgInstAddress + OrgSize > BF->getAddress() + BF->getSize())
+      return createStringError(errc::executable_format_error,
+                               "error reading .altinstructions");
+
+    MCInst *Inst =
+        BF->getInstructionAtOffset(OrgInstAddress - BF->getAddress());
+    if (!Inst)
+      return createStringError(errc::executable_format_error,
+                               "no instruction at address 0x%" PRIx64
+                               " referenced by .altinstructions entry %d",
+                               OrgInstAddress, EntryID);
+
+    // There could be more than one alternative instruction sequences for the
+    // same original instruction. Annotate each alternative separately.
+    std::string AnnotationName = "AltInst";
+    unsigned N = 2;
+    while (BC.MIB->hasAnnotation(*Inst, AnnotationName))
+      AnnotationName = "AltInst" + std::to_string(N++);
+
+    BC.MIB->addAnnotation(*Inst, AnnotationName, EntryID);
+
+    // Annotate all instructions from the original sequence. Note that it's not
+    // the most efficient way to look for instructions in the address range,
+    // but since alternative instructions are uncommon, it will do for now.
+    for (uint32_t Offset = 1; Offset < OrgSize; ++Offset) {
+      Inst = BF->getInstructionAtOffset(OrgInstAddress + Offset -
+                                        BF->getAddress());
+      if (Inst)
+        BC.MIB->addAnnotation(*Inst, AnnotationName, EntryID);
+    }
+  }
+
+  BC.outs() << "BOLT-INFO: parsed " << EntryID
+            << " alternative instruction entries\n";
+
+  return Error::success();
+}
+
 } // namespace
 
 std::unique_ptr<MetadataRewriter>

bolt/test/X86/linux-alt-instruction.s

@@ -0,0 +1,97 @@
+# REQUIRES: system-linux
+
+## Check that BOLT correctly parses the Linux kernel .altinstructions section
+## and annotates alternative instructions.
+
+# RUN: llvm-mc -filetype=obj -triple x86_64-unknown-unknown %s -o %t.o
+# RUN: %clang %cflags -nostdlib %t.o -o %t.exe \
+# RUN:   -Wl,--image-base=0xffffffff80000000,--no-dynamic-linker,--no-eh-frame-hdr,--no-pie
+# RUN: llvm-bolt %t.exe --print-normalized --keep-nops -o %t.out \
+# RUN:   --alt-inst-feature-size=2 | FileCheck %s
+
+## Older kernels used to have padlen field in alt_instr. Check compatibility.
+
+# RUN: llvm-mc -filetype=obj -triple x86_64-unknown-unknown --defsym PADLEN=1 \
+# RUN:   %s -o %t.o
+# RUN: %clang %cflags -nostdlib %t.o -o %t.exe \
+# RUN:   -Wl,--image-base=0xffffffff80000000,--no-dynamic-linker,--no-eh-frame-hdr,--no-pie
+# RUN: llvm-bolt %t.exe --print-normalized --keep-nops --alt-inst-has-padlen \
+# RUN:   -o %t.out | FileCheck %s
+
+## Check with a larger size of "feature" field in alt_instr.
+
+# RUN: llvm-mc -filetype=obj -triple x86_64-unknown-unknown \
+# RUN:   --defsym FEATURE_SIZE_4=1 %s -o %t.o
+# RUN: %clang %cflags -nostdlib %t.o -o %t.exe \
+# RUN:   -Wl,--image-base=0xffffffff80000000,--no-dynamic-linker,--no-eh-frame-hdr,--no-pie
+# RUN: llvm-bolt %t.exe --print-normalized --keep-nops \
+# RUN:   --alt-inst-feature-size=4 -o %t.out | FileCheck %s
+
+## Check that out-of-bounds read is handled properly.
+
+# RUN: not llvm-bolt %t.exe --print-normalized --keep-nops \
+# RUN:   --alt-inst-feature-size=2 -o %t.out
+
+# CHECK:      BOLT-INFO: Linux kernel binary detected
+# CHECK:      BOLT-INFO: parsed 2 alternative instruction entries
+
+  .text
+  .globl _start
+  .type _start, %function
+_start:
+# CHECK: Binary Function "_start"
+.L0:
+  rdtsc
+# CHECK:      rdtsc
+# CHECK-SAME: AltInst: 1
+# CHECK-SAME: AltInst2: 2
+  nop
+# CHECK-NEXT: nop
+# CHECK-SAME: AltInst: 1
+# CHECK-SAME: AltInst2: 2
+  nop
+  nop
+.L1:
+  ret
+  .size _start, .-_start
+
+  .section .altinstr_replacement,"ax",@progbits
+.A0:
+  lfence
+  rdtsc
+.A1:
+  rdtscp
+.Ae:
+
+## Alternative instruction info.
+  .section .altinstructions,"a",@progbits
+
+  .long .L0 - .   # org instruction
+  .long .A0 - .   # alt instruction
+.ifdef FEATURE_SIZE_4
+  .long 0x72      # feature flags
+.else
+  .word 0x72      # feature flags
+.endif
+  .byte .L1 - .L0 # org size
+  .byte .A1 - .A0 # alt size
+.ifdef PADLEN
+  .byte 0
+.endif
+
+  .long .L0 - .   # org instruction
+  .long .A1 - .   # alt instruction
+.ifdef FEATURE_SIZE_4
+  .long 0x3b      # feature flags
+.else
+  .word 0x3b      # feature flags
+.endif
+  .byte .L1 - .L0 # org size
+  .byte .Ae - .A1 # alt size
+.ifdef PADLEN
+  .byte 0
+.endif
+
+## Fake Linux Kernel sections.
+  .section __ksymtab,"a",@progbits
+  .section __ksymtab_gpl,"a",@progbits